Client-server systems split an application into a client part, executed locally on workstations, and a server part, comprising larger scale computations executed on a remote mainframe or parallel computer. In this way, workstations and mainframes can be integrated in a cost-effective and user-friendly way. However, the complexity and scale of such systems is so great that formal quantitative methods are required for their engineering. The proposed project intends to investigate the quantitative design of client-server systems based on stochastic models of performance, with the main emphasis on servers running large database applications - eg ICL's Goldrush Megaserver. We will consider both hardware architecture and system software, in particular database management systems, on both single and multi-processors. By focusing on the key issues of locking, caching, database parallelisation, networks and user response times, the research proposed offers the prospect of a major impact for effective design. The integrated involvement of Metron Technology Ltd, Europe's largest capacity planning enterprise, will facilitate effective validation of models with respect to both simulation and measurement on commercial systems. Moreover, Metron's customised software, eg Athene, will provide a platform for the models' commercial exploitation.